SWE-ATLAS 🤖 Agent Open source

What it does

SWE-ATLAS (Software Engineer AI Agent Atlas) is a template that scaffolds a curated set of skills, subagents, slash commands and engineering conventions into a Claude Code project. Its bet is that Claude Code already handles the build loop (plan mode, goal, auto mode, dynamic workflows), so ATLAS focuses on the front of the loop: deciding what to build and proving the shape works before you spend tokens generating it. Instead of writing heavy specification documents up front, it produces cheap, throwaway artifacts, wireframes and clickable prototypes, so a human and the model can look at the same thing and agree, then hands a clean brief to Claude's native execution.

What is inside

• Slash commands to explore before committing: a low-fi HTML wireframe, a clickable React prototype, a design-variant comparison locked as DESIGN.md, and a visual HTML plan for non-UI changes
• A phase planner that resolves only the load-bearing unknowns through targeted questions, captured as one self-contained HTML document
• Three CLAUDE.md modes: vanilla, collaborative (a review-and-commit loop, the default), and autonomous (approval loop removed, wired for unattended runs)
• A free-will decision skill that, on higher-stakes forks, holds real alternatives open, grounds them in evidence, tries to refute the winner and logs the call in docs/decision_logs/, plus product-owner and UI/UX design skills

Works with

Claude Code. It leans on native Claude Code features (plan mode, /goal, workflows) and can set up Playwright for browser automation.

How to install

npx swe-atlas@latest new-project
npx swe-atlas@latest new-project my-workspace

The CLI walks you through the CLAUDE.md flavour, which skills to install, the design template and optional integrations. A manual git clone --recurse-submodules path also exists.

Maintenance and safety

Published on npm and actively maintained. The autonomous mode removes the approval gate and is built to run unattended, which is powerful but riskier; the collaborative default keeps you in the loop. Decision logs give an audit trail of why choices were made. The repository does not carry a clear open-source licence file, so check terms before reuse.

Who should use it

Developers using Claude Code who want a repeatable discovery-and-design step and senior-engineer conventions, rather than another spec-driven-development framework. Less relevant if you prefer minimal setup or work outside Claude Code.